A method for generating feature maps by using a device adopting CNN including feature up-sampling networks (FPN) is disclosed. The method comprising steps of: (a) allowing, if the input image is obtained, a down-sampling block to acquire a down-sampling image by applying a predetermined operation to an input image (b) allowing, if the down-sampling image is obtained, each of a (1-1)-th to a (1-k)-th filter blocks to acquire each of a (1-1)-th to a (1-k)-th feature maps by applying one or more convolution operations to the down-sampling image and (c) allowing each of up-sampling blocks to receive a feature map from its corresponding filter block, to receive a feature map from its previous up-sampling block, and then rescale one feature map to be identical with the other feature map in size, and to apply a certain operation to both feature maps, thereby generating a (2-k)-th to a (2-1)-th feature maps.